Intelligent Fault Detection and Classification in Semiconductor Manufacturing Using Statistical Learning Techniques

Abstract

Semiconductor manufacturing requires early identification of equipment and process faults to reduce wafer loss and production delay. This study presents an intelligent fault detection and classification framework based on statistical learning techniques. Historical process data were collected from normal and faulty wafer-processing cycles and converted into features describing pressure, gas flow, radio-frequency power, temperature, timing, alarms, and maintenance condition. Linear discriminant analysis, support vector machine, and random forest models were developed and compared. Random forest achieved the highest accuracy of 94.3%, followed by the support vector machine and linear discriminant analysis. Pressure instability showed the strongest classification performance, while timing and equipment faults remained more difficult because they developed gradually and shared features with process drift. The proposed framework can support faster diagnosis, lower false alarms, improved maintenance planning, and more reliable semiconductor production through data-driven monitoring and fault classification.